Invasive non-native species are one of the main causes of degradation of ecosystems worldwide. The control of invasive species is key to reduce threats to ecosystem viability in the long-term. Observations of structural changes in ecological interaction networks following invasive species suppression can be useful to monitor the success of ecological restoration initiatives. We evaluated the structure of plant-bird frugivory interaction networks in a plant community invaded by the guava tree (Psidium guajava L.) by comparing network metrics before and after control actions. Psidium guajava was relevant in all metrics for the unmanaged network in this study, with high degree centrality and high nestedness contribution. Based on the asymmetry of species interactions, we found that birds were highly dependent on the invasive plant before suppression. Once P. guajava trees were eliminated, bird and plant species richness, total number of interactions, and modularity increased, whereas nestedness and interaction strength asymmetry decreased. The diet of the bird community became more diversified once P. guajava was no longer available and relevant species roles in community structure emerged. Our results corroborate the fact that ecological restoration interventions should include the control of non-native plant species that attract frugivorous animals in order to diversify plant-frugivore interactions and thus maintain biodiversity in natural ecosystems.

We conducted the study in the Caieira Natural Municipal Park (CNMP), a protected area that covers 1.27 km² (Joinville, state of Santa Catarina, southern Brazil) (26°18’24”S, 48°47’36”W) (Figure 1a).  The climate is subtropical, constantly humid, with hot summers and no dry season, and average maximum temperatures between 26.0 and 27.6º C (Cfa type in the Köppen classification; Joinville, 2021). The vegetation cover of the CNMP is represented by two forest types: coastal forest (restinga arbórea, in Portuguese) and mangrove (Joinville, 2021), both in the Atlantic Forest domain, an important global biodiversity hotspot (Myers et al., 2000) (Figure 1a,b).  The park grants public access to landscaped areas and trails.

Non-native species were planted in a small orchard in the park between 1970 and 2004. Some of those trees, such as guava (Psidium guajava), avocado (Persea americana), and star fruit (Averrhoa carambola), still bear fruit in the area (Preis, 2020). Several invasive non-native plants have established populations in the park, and are listed in the management plan for control by park managers (Joinville, 2021). An established population of P. guajava trees was present both in landscaped areas and in restinga and mangrove remnants, totaling around 410 individuals per km².

 Psidium guajava is a shrub or tree in the Myrtaceae family, up to 12 m in height, with smooth, scaly, reddish-brown bark. Fruits are globular to pyriform, 2–8 cm long, green to yellow, with pink, yellow, or white pulp (Landrum & Mitra, 2021). Psidium guajava can produce fruits throughout the year, especially in the summer months (Landrum & Mitra, 2021), but fruit phenology varies according to the species variety and water availability (Moura, 2001). In the study area, fruiting occurred between late summer to autumn (February to April; BM, personal observation). The main forms of dispersal are zoochory and anthropochory (Arévalo-Marín et al., 2021; Heleno et al., 2013b). The species is an important node in its introduced range both in pollination networks in Galapagos (see Traveset et al., 2015) and plant–frugivore networks in Galápagos, South Africa and Brazil (see Baltzinger et al., 2020; Heleno et al., 2013b; Silva & Pizo, 2020).

The origin of P. guajava is uncertain, but several authors suggest that it originated between northern South America and Central America (Arévalo-Marín et al., 2021). More recent studies indicate that it may have originated in the savannas and semi-deciduous forests of South America (Landrum & Mitra, 2021) or in the humid Chaco and/or Cerrado (Arévalo-Marín et al., 2021). Psidium guajava is widely cultivated in several countries and considered invasive in several tropical and subtropical regions of the world (Landrum & Mitra, 2021; Richardson & Rejmánek, 2011), such as East Africa (Witt & Luke, 2017), North America (Acevedo-Rodríguez & Strong, 2012), Brazil (Sampaio & Schmidt, 2013; Ziller & Dechoum, 2013) and the Galápagos Islands in Ecuador (Guézou et al., 2010; Urquía et al., 2019). Use by human populations very likely increased its distribution in the past (Arévalo-Marín et al., 2021). As an invasive species, it can form dense stands that displace native vegetation (Leão et al., 2011); thus, the balance between valuable fruit production and invasive potential requires careful monitoring (Landrum & Mitra, 2021) and continuous control.

Data collection of frugivory interactions

We collected data in the CNMP between February and April of 2021 and 2022, when P. guajava trees were fruiting in 2021, and in the same timeframe after P. guajava suppression, in 2022. We carried out thirty field surveys each year. In each survey, the same trail was covered by two observers in early morning hours (6:45 am ± 15 min until 10:00 am ± 30 min) or in the afternoon (4:00 pm ± 15 min and 6:30 pm ± 30 min) (Pizo & Galetti, 2010).  The trails mainly covered landscaped areas and restinga vegetation (Figure 1a). In addition, we defined three observation areas (Figure 1c), where two observers kept watch together for 20 minutes (Jordano & Schupp, 2000). Each observer recorded all avian frugivory events on fruiting plants with a pair of binoculars (Leitz Wetzlar Trinovid 10x40) and a digital camera (Sony DSC-HX300 and Nikon Coolpix p510). Walking direction along the trail and order of observation areas were alternated on sampling days on each field survey to ensure that all study site areas were observed at different times of day. We carried out data collection at the same period of time over two consecutive years to ensure sampling consistency regarding the presence of migratory birds and fructification of the sampled plant species. We performed a nonparametric Wilcox test to verify whether the hours of observation in each year were different between years. The dependent variable was the duration of observation (in minutes) at every fortnight and the independent variable was sampling year. We used the iNEXT function from the iNEXT package to assess community diversity and build species accumulation curves (Hsieh, Ma, & Chao, 2022), using abundance data to estimate bird and plant diversity based on the Shannon index (q = 1). We conducted the analysis using R 3.6.2 software (R core Team, 2022).

We used the following parameters to record frugivory interactions by birds (feeding bouts): the plant species on whose fruits a bird was feeding, and the respective bird species. A new frugivory event was counted if the bird left the tree, and then returned. The number of fruits/seeds ingested per visit was not counted. Bird foraging flocks were recorded as separate interactions per individual. Bird identification was mostly carried out in the field, but we consulted experts and literature in case of uncertainty. Scientific nomenclature follows the Brazilian Committee of Ornithological Records (Pacheco et al., 2021). Likewise, plants not identified in the field were identified with help from experts who reviewed photographic records. The nomenclature of plant species is based on the “Flora e Funga do Brasil” database (2022).

Psidium guajava control campaigns

We conducted two P. guajava control campaigns in the CNMP. In October 2021, control mainly focused on adults and reproductive individuals, but also included small plants in the surroundings of eliminated adults. We eliminated a total of 435 plants. In May 2022, one month after the sampling period, a follow-up expedition took place, in which we eliminated 85 non-reproductive plants as well as resprouted plants. Altogether, we cut down about 520 P. guajava trees in the area. We cut close to ground level with a chainsaw, followed by immediate spraying of a 6% Garlon 480BR (Triclopyr active ingredient) solution on the stump (Dechoum & Ziller, 2013). Once regrowth was observed, we carried out a re-check with foliar application of a 2% Glyphosate solution. During the second field survey, one fruiting P. guajava tree was found, so we eliminated all fruits, the tree was later cut down, and the stump treated.

Network metrics

We organized the interactions observed during each year into quantitative matrices (i=plant species; j=bird species) based on the frequency of frugivory events observed in the CNMP. Two data matrices were built: the first one, entitled “unmanaged community”, referred to the plant community invaded by P. guajava in 2021, prior to the control intervention; the second one, entitled “managed community”, corresponded to the same community after control, in 2022. Both quantitative networks were built and analyzed using the Bipartite package (Dormann et al., 2009) in software R, version 3.6.2 (R core Team 2022). We calculated sampling completeness of species interactions using Chao1 as a richness estimator and the SCw1 function (Macgregor et al., 2017).

We evaluated the following network-level metrics, then contrasted them with those from null models for significance assessments: quantitative nestedness (weighted NODF; Almeida-Neto & Ulrich, 2011), quantitative modularity (DIRTLPAwb+ algorithm; Beckett, 2016), and interaction strength asymmetry (ISA; Dormann et al., 2008). We used Z-scores to assess significance of nestedness and modularity with null-model functions by comparing the results obtained with the observed values, using the methods vaznull and method r2d, respectively, with 999 randomizations each. We also used the null.t.test with 999 randomizations to obtain the significance of interaction strength asymmetry (Dormann et al., 2008). Interaction strength asymmetry varies between −1 and 1, with positive values indicating high dependence of an animal on a plant species, and negative values indicating the opposite (Blüthgen et al., 2007).

We evaluated the following species-level metrics: plant species degree centrality (Martín-González et al., 2010), plant species nestedness contribution, asymmetry of plant species interactions using the push-pull index (Vázquez et al., 2007) and participation coefficients (c-values), and within-module degree (z-values) for network modules (Olesen et al., 2007). Degree centrality takes the number of links of each node (k) into account. The push-pull index indicates the direction of interaction asymmetry based on dependency; positive values indicate that a species more strongly affects the species of the other level with which it interacts than vice versa (“pusher”), while negative values indicate that a species is, on average, more strongly affected by its interaction partners than it affects them (“being pulled”) (Vázquez et al., 2007). The cz-values were evaluated for both plant and bird species. We calculated the thresholds for participation coefficients and within-module degree for each network level (null models with 95% confidence intervals) and used them to identify the topological role of each species (network hub, module hub, connector, or peripheral; Olesen et al., 2007).